Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

The current issue and full text archive of this journal is available on Emerald Insight at:

https://www.emerald.com/insight/1940-5979.htm

RBF
13,1 Investor attention and the response
of US stock market sectors to the
COVID-19 crisis
20 Lee A. Smales
UWA Business School, University of Western Australia, Perth, Australia
Received 26 June 2020
Revised 7 September 2020
26 October 2020 Abstract
Accepted 23 November 2020 Purpose – COVID-19 has had an immense impact on global stock markets, with no sector escaping its effects.
Investor attention towards COVID-19 surged as the virus spread, the number of cases grew and its
consequences imposed on everyday life. We assess whether this increase in investor attention may explain
stock returns across different sectors during this unusual period.
Design/methodology/approach – We adopt the methodology of Da et al. (2015), using Google search
volume (GSV) as a proxy for investor attention to examine the relationship between investor attention and
stock returns across 11 sectors.
Findings – Our results demonstrate that heightened attention towards COVID-19 negatively influences US
stock returns. However, relatively speaking, some sectors appear to have gained from the increased attention.
This outperformance is centred in the sectors most likely to benefit (or likely to lose least) from the crisis and
associated spending by households and government (i.e. consumer staples, healthcare and IT). Such results
may be explained by an information discovery hypothesis in the sense that investors are searching online for
information to enable a greater understanding of COVID-19’s impact on relative stock sector performance.
Originality/value – While we do not claim that investor attention is the only driver of stock returns during
this unique period, we do provide evidence that it contributes to the market impact and to the heterogeneity of
returns across stock market sectors.
Keywords Investor attention, Stock market returns, Google search volume, COVID-19, Coronavirus
Paper type Research paper

1. Introduction
The economic and social impact of coronavirus (COVID-19) has been substantial and has
captured the attention of all. As a health crisis has mutated into a financial crisis, retail
investors have suffered substantial losses in quick time [1], risk has increased substantially
(Zhang et al., 2020) and the level of market volatility reached virtually unprecedented heights
(Baker et al., 2020). As the number of confirmed cases has grown, markets have witnessed a
significant decline in liquidity (Baig et al., 2020) and value, with a response greater than that to
any other public health emergency in at least the prior decade (Schell et al., 2020). The largest
impact was initially felt in Chinese stocks (Al-Awadhi et al., 2020), US stocks with
international exposure (Ramelli and Wagner, 2020) and in countries that suffered from the
2003 SARS outbreak (Ru et al., 2020) but subsequently spread across the globe (Ali et al., 2020)
along with negative media sentiment (Haroon and Rizvi, 2020). While global markets had
recovered a substantial proportion of losses suffered at the depth of the crisis (mid-March
2020) by the end of our sample period (end-May 2020), it is not at all clear how persistent the
economic fallout will be, or when markets will regain some sense of normality.
We contribute to the discussion regarding the impact of COVID-19 on financial markets
through the prism of investor attention. Investor attention refers to the cognitive constraints
(Kahneman, 1973) that investors face owing to the large amount of information available to
them (an issue that is more prevalent than ever in the age of the Internet and social media).
Review of Behavioral Finance
This means that investors must be selective in processing information (Peng and Xiong, 2006)
Vol. 13 No. 1, 2021 and can lead to both underreaction due to a delayed response to key information and
pp. 20-39
© Emerald Publishing Limited
1940-5979
DOI 10.1108/RBF-06-2020-0138 JEL Classification — G01, G10, G14
overreaction when too much attention is paid to irrelevant information (Lim and Teoh, 2010; Investor
Barber and Odean, 2013; Hirshleifer, 2015). attention
Two principal theories help to explain why investor attention influences stock returns:
both are related to the idea that attention is a scarce resource. The first, conceived by Odean
during the
(1999) and Barber and Odean (2008), is based on the premise of asymmetric choice. That is, COVID-19 crisis
when retail investors are seeking to buy stocks, they do not have enough resources to devote
to the entire universe of feasible investments and so create price pressure by focussing
purchases on stocks that have grabbed their attention. Conversely, when looking to sell 21
stocks, retail investors are typically only able to see stocks they already own. The same
limitations do not apply to institutional investors since they can shorten stocks, have access
to greater resources and have the ability to apply screening techniques to the investment
universe. The alternate theory contends that investor attention leads to information discovery
and so creates price pressure (Vlastakis and Markellos, 2012) and improves market efficiency
(Vozlyublennaia, 2014). This second theory is also consistent with the notion that prices only
respond to new information when investors pay attention to it (Huberman and Regev, 2001).
Our analysis seeks to go some way towards empirically resolving the tension between these
competing theories, and our results seem to favour the information discovery hypothesis.
Da et al. (2011) offer a number of indirect proxies for investor attention but suggest that
Google Search volume is a direct measure with several key advantages. Following their lead,
we use Google Search Volume (GSV) as a proxy for investor attention in order to investigate
whether particular stock market sectors are likely to benefit (or suffer) from a greater amount
of investor attention during the crisis. That is, our empirical analysis seeks to address a
relatively simple, but important, research question, namely, does investor attention influence
returns across stock sectors during the COVID-19 pandemic? If so, is the effect
heterogeneous?
Several studies have previously adopted GSV to examine the effect of investor attention
on financial markets. This includes analysis of stock markets in North America (Da et al.,
2011, 2015; Ding and Hou, 2015; Bijl et al., 2016; Tang and Zhu, 2017), Europe (Bank et al.,
2011; Aouadi et al., 2013; Kim et al., 2019) and Asia (Takeda and Wakao, 2014; Tantaopas
et al., 2016). When investors are paying attention (GSV increases), more information flows
into the marketplace (Smith, 2012) and is incorporated into prices more quickly (Andrei and
Hasler, 2014) leading to more trading activity (Preis et al., 2010; Bank et al., 2011; Takeda and
Wakao, 2014; Kim et al., 2019) and higher volatility (e.g. Vlastakis and Markellos, 2012;
Aouadi et al., 2013; Dimpfl and Jank, 2015; Goddard et al., 2015; Kim et al., 2019).
Generally, a spike in GSV is associated with a contemporaneous market response that is
subsequently reversed (Da et al., 2011, 2015; Tang and Zhu, 2017; Heyman et al., 2019; Smales,
2020) unless new information is forthcoming (Cheng et al., 2019). Vozlyublennaia (2014)
argues that this response to investor attention improves market efficiency by making
markets less predictable. However, there is no consensus about the direction of the influence
of GSV on stock returns. The extant literature provides evidence of a positive influence (Bank
et al., 2011; Da et al., 2011; Tang and Zhu, 2017), a negative influence (Vozlyublennaia, 2014;
Da et al., 2015; Bijl et al., 2016; Chen, 2017) and no influence (Preis et al., 2010; Takeda and
Wakao, 2014; Kim et al., 2019) on returns. We contribute to this discussion by providing
additional evidence of a negative relationship.
To-date, the empirical investigation has ignored stock market sectors and focussed on
market indexes (Vozlyublennaia, 2014; Chen, 2017) and individual firms (e.g. Da et al., 2011;
2015; Takeda and Wakao, 2014). Examination of the heterogeneity of the GSV-return
relationship is limited. For instance, Da et al. (2011) suggest the relationship is stronger for
small stocks and those traded more frequently by retail investors, while Da et al. (2015) show
the importance of stocks with higher beta, higher volatility and greater downside risk (such
stocks are also likely to be small). Neither investigate whether the relationship is related to the
RBF stock sector in which firms operate. We attempt to rectify this gap in the research by
13,1 examining the relative impact of GSV across different stock market sectors.
While our sample period is relatively short, running from December 2019 to the end of
May 2020, it offers a unique setting in which the world’s media, financial markets and general
population were primarily focussed on a single theme – “coronavirus” – and that
encompasses a period of extreme market stress. It is interesting to study such periods
since they are associated with heightened sensitivity to news (Garcia, 2013) and a greater
22 sense of uncertainty triggered by anxiety and sadness (Smith and Ellsworth, 1985).
Our main analysis focuses on returns in US stock sectors (we use the 11 GICS defined
sectors). As noted, the literature has focussed largely on general market indexes and the
aggregate effect of investor attention on individual stocks. In focussing on sectors we are able
to provide a closer examination as to the heterogeneous impact of investor attention. This is
particularly important during periods such as the coronavirus pandemic because there is a
great potential for some sectors to benefit (e.g. those providing services for people working
from home) to the detriment of others (e.g. international travel and tourism). We focus on US
stocks because it has the largest and most liquid stock markets in the world, accounting for
approximately 50% of global market capitalization [2]. As a robustness check, and an
indication as to the generalizability of our US results, our analysis concludes with a brief
study of global sectorial returns (based on MSCI sector indexes).
In summary, our contribution is three-fold; first, we add to the discussion as to the
direction of the GSV-return relationship and contribute empirical evidence to the debate on
the theory of investor attention. Second, we provide a greater understanding as to the
heterogenous nature of the GSV-return relationship. Finally, we believe we are the first to
examine this particular relationship in the context of the extreme market events of the
COVID-19 pandemic.
The information discovery hypothesis seems to offer a better rationale for the results
portrayed in our empirical analysis, in the sense that retail investors are using Google searches so
as to gain an understanding of coronavirus and its effect on relative stock returns across different
sectors. We show that while heightened investor attention (increase in GSV) towards coronavirus
is associated with negative stock returns in general, there is relative outperformance in the sectors
most likely to benefit from related changes in household and government spending (i.e. consumer
staples, healthcare and IT). These sectors may be thought of as “essential”, and to some extent they
also benefit from relatively stable government spending. Interestingly, we do not find a
relationship between the reported number of cases and deaths related to COVID-19 and stock
returns. This suggests that investors may be concentrating on forward-looking information
rather than current levels of COVID-19 cases. We do not claim that investor attention is the only
driver of stock returns during the crisis, but we do provide evidence that it contributes to the
heterogeneity of returns across stock market sectors.

2. Data and methodology


2.1 Stock returns
US stock returns form the dependent variable in the main part of our study. In addition to
using the S&P500 Composite Index as a measure of overall market performance, we consider
returns on the 11 sectors [3] within the Global Industry Classification Standard (GICS) with
data for the relevant S&P Indexes taken from DataStream. In the latter part of our study, we
refer to global markets and utilise the MSCI World Index and associated sector indexes. In
both cases, we compute two time series of daily log returns. The first is a set of returns in
excess of the risk-free 1M T-Bill rate, and the second is a set of returns adjusted for the T-Bill
rate and the market (S&P500) return. Our sample period runs from 31 December 2019 to 31
May 2020, providing a sample of 104 trading days.
Table 1 provides descriptive statistics for both sets of stock index returns. The energy Investor
sector provided the most negative and most volatile returns in both US and world markets. attention
Conversely, IT was the top performing sector, possibly relating to the increased take-up of
video conferencing and media streaming services during lockdown. All sectors exhibit
during the
negative skewness and “fat-tails” typical of asset returns. COVID-19 crisis
Figure 1 depicts the performance of relative sectors during the COVID-19 crisis. It shows
the market trough occurred in mid-March, shortly after the US declared a national
emergency, and also reflects the relative underperformance of the energy sector 23

2.2 Google Search Volume (GSV)


Our study uses GSV as a proxy for investor attention. Searching for information about a
specific subject directly demonstrates that one is paying attention to that subject, so GSV
provides a direct and timely measure of information acquisition. The dominance of Google in
the search query market suggests that GSV will be representative of the search behaviour of
the general population. Since institutional investors can access a number of information
sources, it is most likely that this is a measure of the attention by retail investors (Da et al.,
2015; Ding and Hou, 2015), although it may also offer a secondary source of information for
institutional investors.
The GSV index, obtained via Google Trends [4], indicates the volume of search queries on
a particular topic in proportion to all searches within a particular location and time period.
GSV is standardized to lie within a range of 0–100, where 100 indicates a particularly active
search query. Google Trends applies filters to remove duplicate searches and searches
occurring very infrequently. The relatively short sample period of our study means that we
can form a daily GSV time series without the normalization process required for longer
periods [5]. For analysis of US stock returns, we focus on GSV for US searches, and for world
stock returns we focus on global GSV.
Our study primarily focuses on searches for the “coronavirus” keyword [6]. This keyword
is chosen as it is specific to our sample period (unlike “epidemic” or “pandemic”) and
circumstance (unlike “virus” which may also relate to computers). During our sample period,
search volume for “coronavirus” is also substantially higher than for related terms such as
“COVID”. For instance, during March, GSV for “coronavirus” averaged 66.8, as opposed to
12.8 for “COVID”. Table 1 panel A shows that the daily changes in GSV are generally positive,
exhibit positive skewness and a large degree of kurtosis.
Figure 2 illustrates US and global GSV during our sample period. GSV increases
substantially from late-February, peaks in mid-March (as stock markets bottom) and then
declines substantially as stock markets recover. The largest change in GSV occurs on 12
March which is the day after the US blocked European travel and immediately preceded the
declaration of a national emergency.

2.3 COVID-19– related cases and deaths


In addition to GSV, we consider whether news directly related to the COVID-19 outbreak
influences market returns. We obtain information on the daily number of new US cases and
deaths from the European Centre for disease Prevention and Control (ECDC) [7]. The ECDC
collects and harmonizes data related to the pandemic from around the world, providing a
global perspective [8]. During our sample period, Table 1 shows the daily average of new
cases was over 10,900 with COVID-19–related deaths averaging more than 683.
Figure 3 shows the daily number of reported COVID-19 cases, and related deaths in the US
cases started to increase towards the end of February 2020, with reported deaths lagging by
around two weeks. Both reported cases and related deaths peaked in April 2020.
24
13,1
RBF

Table 1.
Descriptive statistics
Level First difference / Return
Mean Std dev Mean Std dev Min Max Skewness Kurtosis Jarque-Bera

Panel A: Google Search Volume (GSV) þ COVID-19 variables


GSV(US) 25.0 24.5 0.078 5.71 15.00 42.00 3.67 30.05 3,371.7
GSV(Global) 30.8 27.0 0.126 4.96 14.00 30.00 2.21 15.80 787.5
Cases (’000) 10.9 12.3 0.212 3.39 19.70 9.22 2.01 14.29 616.7
Deaths 683.2 950.4 11.41 533.10 2629 2520 0.38 13.23 451.9
Panel B: US stock sectors (S&P)
S&P500 2,972.0 300.7 0.066 3.13 12.77 8.97 0.51 6.56 58.5
Communications 171.9 15.6 0.017 2.78 11.03 8.80 0.55 6.40 54.4
Consumer discretionary 928.4 90.7 0.003 2.92 12.88 8.29 1.00 7.55 104.7
Consumer staples 612.0 39.0 0.056 2.56 9.69 8.07 0.06 6.41 49.5
Energy 324.7 86.1 0.448 4.97 22.42 15.11 0.92 7.53 101.5
Financials 420.3 73.7 0.281 4.04 15.07 12.43 0.26 5.77 33.8
Healthcare 1134.5 79.2 0.006 2.73 10.53 7.31 0.20 5.48 26.8
Industrials 591.8 92.2 0.201 3.49 12.16 12.00 0.25 5.69 31.8
IT 1594.0 136.1 0.046 3.55 14.98 11.30 0.41 6.86 66.1
Materials 334.3 38.9 0.088 3.37 12.15 11.00 0.40 5.48 28.8
Real Estate 220.7 25.5 0.099 3.70 18.09 8.28 1.02 7.90 119.7
Utilities 310.2 31.4 0.068 3.60 12.27 12.32 0.00 5.71 31.2
Panel C: Global stock sectors (MSCI)
MSCI World 507.3 55.6 0.092 2.50 10.00 8.06 0.90 7.45 98.7
Communications 77.6 6.6 0.013 2.23 9.00 5.55 0.95 7.19 90.8
Consumer discretionary 255.1 26.6 0.031 2.45 9.89 7.86 0.98 7.72 112.4
Consumer staples 235.4 16.5 0.070 1.90 9.25 4.87 0.91 8.26 133.2
Energy 154.4 37.7 0.400 4.14 19.95 13.94 1.27 9.49 208.3
Financials 107.2 19.1 0.285 3.07 11.44 9.89 0.60 6.73 66.0
Healthcare 270.3 18.6 0.017 2.20 8.34 6.27 0.49 6.03 43.5
Industrials 226.7 32.8 0.164 2.64 10.42 9.10 0.49 6.73 63.7
IT 306.5 26.9 0.042 2.98 12.51 9.30 0.58 7.13 79.0
Materials 237.3 29.5 0.122 2.55 10.94 9.28 0.83 7.78 109.7

(continued )
Level First difference / Return
Mean Std dev Mean Std dev Min Max Skewness Kurtosis Jarque-Bera

Real estate 189.2 25.0 0.169 2.80 13.49 7.63 1.21 8.30 145.8
Utilities 140.0 14.3 0.084 2.72 11.57 7.93 0.56 6.87 69.9
Panel D: Control variables
VIX 33.1 18.0 0.671 11.23 26.62 38.22 1.20 5.21 45.5
EPU 315.7 208.8 3.623 94.05 309.0 358.3 0.27 5.66 31.6
ADS 14.6 11.5 0.163 0.61 2.83 1.47 1.03 6.94 84.7
TERM 0.4 0.2 0.001 0.06 0.18 0.28 1.01 7.47 103.4
CSPRD 1.2 0.4 0.005 0.08 0.30 0.47 1.40 16.25 786.8
Note(s): This table shows descriptive statistics for the data used in this study. Level and first difference data for Google Search Volume (GSV) for the “coronavirus” key
word and the daily number of US COVID-19 identified cases and deaths are shown in panel A. Level and returns for US stock sectors are shown in panel B and for global
stock sectors in Panel C. Panel C shows summary information for the set of control variables; the CBOE VIX index (VIX), the US economic policy uncertainty index (EPU),
the Aruoba–Diebold–Scotti business conditions index (ADS), the 1–10 years term premium (TERM), and the spread between corporate bonds yielded Baa and Aaa
(CSPRD). Sample period: 31-Dec-2019–31-May-2020
25
during the
attention
Investor

Table 1.
COVID-19 crisis
RBF 120
13,1

100

26
80

60

40

S&P500 Comms. Cons. Disc. Cons. Staples


20 Energy Financials Healthcare Industrials

Figure 1. I.T. Materials Real Estate Utilities


US stock returns
(rebased) for different
sectors 0
Dec-19 Jan-20 Jan-20 Feb-20 Feb-20 Mar-20 Mar-20 Apr-20 Apr-20 May-20 May-20

120

100

80

60
GSV(US)

40 GSV(Global)

20
Figure 2.
Google Search Volume
for “Coronavirus”
search term 0
Dec-19 Jan-20 Jan-20 Feb-20 Feb-20 Mar-20 Mar-20 Apr-20 Apr-20 May-20 May-20
40000 6000 Investor
attention
35000
5000 during the
30000
COVID-19 crisis
4000
25000 27
20000 3000
Cases
15000
Deaths 2000
10000
1000
5000 Figure 3.
Number of daily US
coronavirus cases /
0 0 deaths
Dec-19 Jan-20 Feb-20 Mar-20 Apr-20

2.4 Controlling for other news types


The main hypothesis of our study is that investor attention is influential in determining stock
returns during the COVID-19 crisis. Of course, it is possible that in addition to Internet
searches, investors are considering other types of news. To account for this we follow Da et al.
(2015) and Han et al. (2018) and introduce a set of control variables that relate to financial
market conditions (CBOE Implied Volatility Index – VIX), macroeconomic developments (the
Aruoba-Diebold-Scotti business conditions index – ADS) and uncertainty around
government policy decisions (the economic policy uncertainty index – EPU). We further
control for economic conditions using measures for the term structure (difference between
yields on 1 and 10 years Treasuries – TERM) and credit spreads (difference between the yield
on corporate bonds rated Baa and Aaa by Moodys – CSPRD).
Table 2 provides a correlation matrix for the variables used in the study. We see a first
indication that GSV and stock returns are negatively related (0.48), and global Internet
searches are correlated with those in the US (0.85). As one might expect, since one naturally
follows the other, the number of deaths is positively related to the number of cases (0.96).
ΔVIX is significantly negatively correlated with stock returns (consistent with the prior
signed literature, e.g. Whaley, 2009) and positively related with our measure of investor attention
(ΔGSV) – this creates a possible issue with multicollinearity that we discuss in our empirical
results.

2.5 Methodology
Our empirical analysis adopts the methodology of Da et al. (2015) and Han et al. (2018). The
main regression specification is of the following form:
X
rt ¼ β0 þ β1 ΔGSVt þ β2 Casest þ β3 rt−1 þ γ m Controlm;t þ εt (1)
m

where rt denotes the return for a stock market sector index on day t, and rt-1 is the lagged
return. ΔGSVt is the daily change (first difference) in the Google Search Volume of the
“coronavirus” key word. Casest is the natural log of the daily number of new US cases of
28
13,1
RBF

Table 2.
Correlation matrix
rSP500 ΔGSV(US) ΔGSV(Global) Cases Deaths ΔVIX ΔEPU ΔADS ΔTERM

ΔGSV(US) 0.476
ΔGSV(global) 0.490 0.853
Cases 0.083 0.161 0.219
Deaths 0.155 0.221 0.267 0.956
ΔVIX 0.740 0.442 0.453 0.167 0.204
ΔEPU 0.125 0.146 0.035 0.002 0.032 0.131
ΔADS 0.284 0.199 0.233 0.215 0.373 0.247 0.053
ΔTERM 0.324 0.096 0.011 0.021 0.021 0.213 0.060 0.121
ΔCSPRD 0.105 0.032 0.020 0.008 0.020 0.082 0.019 0.323 0.115
Note(s): This table shows the Pearson correlation coefficients for the variables used in this study. This includes the daily S&P500 return (in excess of the 1M T-Bill rate),
the daily change in US and global Google Search Volume (GSV) for the “coronavirus” key word, the daily number of US COVID-19 identified cases and deaths and daily
changes in the CBOE VIX index (VIX), the US economic policy uncertainty index (EPU), the Aruoba–Diebold–Scotti business conditions index (ADS), the 1–10 years term
premium (TERM) and the spread between corporate bonds yielded Baa and Aaa (CSPRD). Significance at the 5% level (or better) is indicated by italics. Sample period: 31-
Dec-2019–31-May-2020
COVID-19 (because of the extremely high correlation between cases and deaths, we do not use Investor
both variables in the same model). The set of control variables includes daily changes in the attention
CBOE implied volatility index (ΔVIX), a news-based measure of economic policy uncertainty
(ΔEPU), changes in the Aruoba–Diebold–Scotti (ΔADS) business conditions index, changes
during the
in the 1–10 years term structure (ΔTERM) and the difference between the yield on corporate COVID-19 crisis
bonds rated Baa and Aaa by Moodys (ΔCSPRD). εt are Newey–West standard errors.
29
3. Empirical analysis
3.1 Baseline model
Initially, we use excess returns adjusted only for the risk-free rate (unadjusted for general
market returns) as our dependent variable. The estimated coefficients reported in Table 3 signed
show that GSV is negatively related to US stock returns for the S&P 500 and for all of the
individual sectors. This implies that there is a decline in stock returns as retail investors
increase their “googling” for coronavirus information. The results are economically and
statistically significant in the univariate case (Panel A), after controlling for COVID-19 cases
in the bivariate case (panel B) and after controlling for other factors (panel C). In the univariate
model, the largest (weakest) effect of GSV is on the energy (consumer staples / healthcare)
sector. However, after including the full set of control variables, the effect of a 1 SD increase in
GSV ranges from 2.17% (0.089 3 24.5) for healthcare stocks to 4.21% for the utilities
sector. While an increase in the number of cases tends to have a negative impact on stock
returns, the effect is not statistically significant.
An increase in the number of COVID-19 cases is also associated with lower stock returns signed
across all sectors, however, this relationship is not statistically significant. The estimated
coefficients for the control variables align with ex ante expectations. Stock returns are lower
when uncertainty is higher (ΔVIX and ΔEPU) and higher when business conditions improve
(ΔADS). A 1 SD increase in VIX is associated with a 2.56% (0.142 3 3 18.0) return for the
S&P500, a 3.94% return for the energy sector and just 1.60% return for utilities. Since the
yield curve typically steepens when monetary policy is eased, the positive coefficient for
ΔTERM is consistent with returns responding positively to Fed policy action. The negative
coefficients for lagged returns are consistent with negative autocorrelation often associated
with daily stock returns. In all cases, the model explains a large proportion of the variance in
returns. Together, the information in Table 3 indicates that investor attention provides an
influence on stock returns in addition to the other news released during the pandemic. The
results for the baseline model are almost identical across panels A (ΔGSV only) and B (ΔGSV
and cases), and it is only when we add the additional variables, particularly ΔVIX, in panel C
that the results change in any meaningful way. In this case, the magnitude of the ΔGSV
coefficient declines – likely because ΔVIX contains some of the same information for stock
price returns.
The identified negative GPR-return relationship suggests that, consistent with Da et al.
use this in
(2015), investors are using Google to acquire information to resolve household concerns report
during the crisis – consistent with the information discovery hypothesis. However, the
magnitude of the response is greater than that identified by Da et al. (2015); this could be due
to their use of a constructed index rather than a specific search term or the greater impact of
investor attention during the period of the COVID-19 crisis. The magnitude of estimated
results for the S&P500 index is similar to that found in Smales (2020). The greater (smaller)
magnitude of coefficients for the energy and financial (consumer staples and healthcare)
sectors are also consistent with the stronger relationship in stocks with higher beta noted by
Da et al. (2015). The sectors that relatively outperform (suffer the least) are those that may
appear to provide essential services (food and healthcare) and potentially benefit from stable
spending patterns during a downturn.
30
13,1
RBF

Table 3.

returns (by sector)


Regression: Influence

excess (Rs–Rf) equity


of investor attention on
S&P500 Comms Cons. Disc Cons. Staples Energy Financials Healthcare Industrials IT Materials Real estate Utilities

Panel A: Univariate
Constant 0.038 (0.196) 0.015 (0.162) 0.036 (0.203) 0.048 (0.179) 0.409 (0.379) 0.246 (0.273) 0.024 (0.190) 0.161 (0.259) 0.085 (0.206) 0.078 (0.232) 0.093 (0.304) 0.061 (0.272)
ΔGSV(US) 0.260*** (0.043) 0.241*** (0.039) 0.257*** (0.039) 0.206*** (0.057) 0.340*** (0.056) 0.299*** (0.051) 0.211*** (0.036) 0.265*** (0.047) 0.286*** (0.049) 0.273*** (0.038) 0.238*** (0.055) 0.264*** (0.054)
Adj. R2 0.219 0.238 0.246 0.205 0.145 0.172 0.189 0.181 0.205 0.208 0.127 0.169
F-Statistic 29.68 32.87 34.19 27.36 18.32 22.17 24.70 23.49 27.34 27.85 15.86 21.74

Panel B: Bivariate
Constant 0.063 (0.185) 0.024 (0.187) 0.086 (0.186) 0.036 (0.175) 0.836** (0.343) 0.256 (0.222) 0.052 (0.178) 0.175 (0.201) 0.125 (0.239) 0.232 (0.200) 0.022 (0.210) 0.139 (0.218)
ΔGSV(US) 0.259*** (0.045) 0.239*** (0.040) 0.254*** (0.041) 0.208*** (0.058) 0.329*** (0.058) 0.299*** (0.053) 0.209*** (0.038) 0.264*** (0.048) 0.287*** (0.051) 0.270*** (0.040) 0.241*** (0.056) 0.269*** (0.054)
Cases 0.005 (0.036) 0.007 (0.032) 0.022 (0.039) 0.015 (0.033) 0.076 (0.057) 0.002 (0.050) 0.014 (0.034) 0.002 (0.045) 0.007 (0.040) 0.028 (0.044) 0.021 (0.055) 0.035 (0.050)
Adusted. 0.211 0.231 0.239 0.198 0.142 0.164 0.181 0.172 0.197 0.202 0.119 0.163
R2
F-statistic 29.68 16.29 17.03 13.60 9.41 10.97 12.27 11.63 13.54 13.90 7.89 10.90

Panel C: Multiple
Constant 0.351** (0.164) 0.314** (0.157) 0.336* (0.178) 0.246 (0.161) 0.131 (0.316) 0.327 (0.223) 0.225 (0.170) 0.389* (0.202) 0.564*** (0.189) 0.282 (0.189) 0.717** (0.324) 0.511* (0.299)
ΔGSV(US) 0.118*** (0.037) 0.111*** (0.032) 0.128*** (0.032) 0.115* (0.062) 0.103** (0.050) 0.127*** (0.044) 0.089** (0.036) 0.112** (0.047) 0.114*** (0.035) 0.130*** (0.038) 0.088** (0.040) 0.172*** (0.059)
Cases 0.037 (0.042) 0.025 (0.034) 0.022 (0.045) 0.040 (0.037) 0.022 (0.072) 0.060 (0.059) 0.013 (0.042) 0.057 (0.050) 0.050 (0.042) 0.024 (0.049) 0.093 (0.072) 0.075 (0.065)
ΔVIX 0.142*** (0.024) 0.127*** (0.019) 0.139*** (0.023) 0.102*** (0.028) 0.219*** (0.045) 0.172*** (0.036) 0.115*** (0.023) 0.149*** (0.031) 0.173*** (0.022) 0.145*** (0.027) 0.149*** (0.039) 0.089** (0.040)
ΔEPU 0.007 (0.018) 0.003 (0.017) 0.017 (0.016) 0.010 (0.020) 0.016 (0.041) 0.006 (0.023) 0.010 (0.020) 0.001 (0.023) 0.012 (0.017) 0.017 (0.020) 0.012 (0.027) 0.001 (0.035)
ΔADS 0.792 (0.533) 0.679 (0.421) 0.710 (0.586) 0.114 (0.508) 2.123** (0.808) 1.227* (0.673) 0.434 (0.559) 1.178* (0.658) 0.698 (0.517) 0.929 (0.559) 1.313 (0.880) 0.770 (0.713)
ΔTERM 7.556* (4.388) 4.523 (3.031) 3.484 (3.929) 9.079** (4.240) 1.979 (2.887) 7.298 (6.985) 9.406** (4.039) 8.045 (6.271) 8.280** (4.096) 9.569* (5.089) 9.645* (5.429) 14.375** (6.156)
ΔCSPRD 3.938 (2.454) 4.675* (2.384) 5.229** (2.408) 1.917 (2.645) 5.857 (3.989) 3.784 (3.275) 1.020 (2.667) 5.508* (3.294) 5.355** (2.547) 4.124 (2.688) 4.840 (3.339) 4.238 (4.062)
return(1) 0.219*** (0.079) 0.257*** (0.046) 0.142 (0.087) 0.171 (0.108) 0.081 (0.087) 0.185* (0.100) 0.188** (0.077) 0.061 (0.113) 0.267*** (0.055) 0.142 (0.099) 0.106 (0.143) 0.151 (0.144)
Adj. R2 0.639 0.677 0.595 0.514 0.439 0.513 0.554 0.481 0.694 0.547 0.428 0.362
F-Statistic 23.32 27.52 19.59 14.36 10.90 14.31 16.65 12.68 29.59 16.23 10.45 8.16
Note(s): This table reports the estimated coefficients for the regression specified in Eqn (1). The dependent variable is the daily return on the S&P500 Composite index or
one of the 11 S&P sector indexes (in excess of the risk-free 1M T-Bill rate). The key explanatory variables are the daily change (first difference) in the Google Search Volume
(ΔGSV) for “coronovirus” in the US and the natural log of the daily number of new US cases of COVID-19. Control variables include the CBOE implied volatility index
(VIX), economic policy uncertainty (EPU), the Aruoba–Diebold–Scotti business conditions index (ADS), the 1–10 years term premium (TERM) and the credit spread
between corporate bonds yielded Baa and Aaa (CSPRD). Newey–West standard errors are shown in parentheses. ***, ** and * indicate statistical significance at the 1, 5
and 10% level respectively. Sample period: 31-Dec-19–31-May-20
3.2 Effect on short-term investment horizon Investor
We follow Da et al. (2015) in examining the influence of investor attention on the structure of attention
stock returns for investment horizon tþk. We report the results for k 5 0 to 5 in Table 4. In
part, our results are consistent with Da et al. (2015) in that we find some evidence of a short-
during the
term reversal. They differ to the extent that the reversal (k 5 1) is only partial, and investor COVID-19 crisis
attention then appears to have an impact for several days afterwards (suggestive of
momentum). That is, investor attention appears to have a lasting influence on stock returns,
with a 1 SD increase in ΔGSV reducing S&P500 returns by 4.70% over five days. 31

3.3 Influence of investor attention on sectors relative to overall market


Since the COVID-19 crisis has had a market-wide impact, it makes sense to also assess
whether investor attention (GSV) influences returns that are adjusted for returns in the
overall index. This allows an assessment as to which, if any, sectors perform better than the
general market when investor attention towards “coronavirus” is heightened. We repeat our
earlier analysis with adjusted returns ðrSector;t − ðrf ;t þ rS&P500;t ÞÞ as our dependent variable
and report the results in Table 5.
The bivariate results (panel A) show that, relatively speaking, the consumer staples and
healthcare sectors benefited from increased investor attention. A 1 SD increase in GSV
induces an excess return of 1.23% (0.05 * 24.5) in both sectors. Conversely, the energy,
financial, and IT sectors have a negative relationship with ΔGSV. For the energy sector, a 1
SD increase in GSV is associated with an excess return of 1.74%. The sectors that appear to
benefit from investor attention are also those sectors which one might expect to benefit most
from COVID-19, for instance, via increased stockpiling of food and sanitary products or
provision of medication and medical equipment. Consistent with the information discovery
hypothesis, this would suggest that retail investors are using Google searches so as to gain an
understanding of COVID-19 and the influence it may have on economic and financial
performance of specific sectors. Again, there is little evidence that the number of COVID-19
cases affects stock returns in a statistically significant way.
As we mention earlier, ΔVIX is significantly correlated with both SP500 returns (which now
constitute part of the dependent variable and its lag) and our key explanatory variable, ΔGSV.
This creates the real possibility that including ΔVIX in the regression model will introduce
multicollinearity and so unintentionally introduce error into our coefficient estimates. To explore
this issue, we report estimates for models including (panel B) and excluding (panel C) ΔVIX.
When we incorporate the control variables other than ΔVIX (panel B), the results are
similar to the bivariate model. That is, relative to the overall market, consumer staples and
healthcare benefit from increased investor attention, while the energy, financial and IT sectors
suffer. When we also add ΔVIX, the only significant ΔGSV relationship that remains is for the
Healthcare sector – with positive relative outperformance. The reduction in statistically
significant relationships for ΔGSV may be due to ΔVIX subsuming the explanatory power of
ΔGSV (i.e. they incorporate the same / similar information), indicated by the high degree of
correlation (∼0.45), or it may be due to the statistical issues already identified. Regardless, it is
clear that ΔGSV has a positive influence on the performance of the healthcare sector.

3.4 Causality
We next conduct a set of Granger (1969) causality tests in order to explore the direction of
causality. To some extent, the causality tests also serve as a sensibility test for our analysis.
For instance, we should not expect stock returns or investor attention to cause an increase in
the number of new cases, and we do not. We do find evidence to support our results since we
can reject the hypothesis that investor attention does not cause stock returns (or ΔVIX) while
rejecting the reverse causality. Our results are consistent with the US results of
32
13,1
RBF

returns
Table 4.

structure of S&P500
Regression: Influence
of investor attention on
S&P500 S&P500 (þ1) S&P500 (þ2) S&P500 (þ3) S&P500 (þ4) S&P500 (þ5)

Constant 0.351** (0.164) 0.569 (0.386) 0.351** (0.164) 0.289 (0.380) 0.504 (0.455) 0.395 (0.567)
ΔGSV(US) 0.118*** (0.037) 0.101 (0.070) 0.118*** (0.037) 0.129** (0.050) 0.245*** (0.060) 0.261***(0.068)
Cases 0.037 (0.042) 0.089** (0.044) 0.037 (0.038) 0.002 (0.084) 0.026 (0.102) 0.007 (0.125)
Adj. R2 0.639 0.160 0.699 0.298 0.481 0.437
F-Statistic 23.32 3.38 30.41 6.31 12.47 10.53
Note(s): This table reports the estimated coefficients for the regression specified in Eqn (1). The dependent variable is the return on the S&P500 Composite Index (in
excess of the 1M T-Bill risk-free rate) between day t and day tþk. The key explanatory variable is the daily change (first difference) in the Google Search Volume (ΔGSV)
for “coronovirus” in the US and the natural log of the daily number of new US cases of COVID-19. Control variables are included in the regression but not reported. Newey–
West standard errors are shown in parentheses. ***, **, and * indicate statistical significance at the 1, 5,and 10% level, respectively. Sample period: 31-Dec 2019–31-
May 2020
Comms Cons. Disc Cons. Staples Energy Financials Healthcare Industrials IT Materials Real estate Utilities

Panel A: Bivariate
Constant 0.039 (0.143) 0.023 (0.141) 0.099 (0.208) 0.773 (0.454) 0.193 (0.222) 0.017 (0.152) 0.112 (0.172) 0.188 (0.127) 0.169 (0.159) 0.085 (0.231) 0.198 (0.274)
ΔGSV(US) 0.019 (0.016) 0.005 (0.016) 0.051** (0.023) 0.071** (0.031) 0.039* (0.021) 0.050*** (0.017) 0.005 (0.019) 0.026** (0.012) 0.011 (0.018) 0.018 (0.026) 0.010 (0.031)
Cases 0.002 (0.020) 0.017 (0.020) 0.019 (0.029) 0.072 (0.063) 0.003 (0.031) 0.009 (0.021) 0.002 (0.024) 0.012 (0.018) 0.023 (0.022) 0.025 (0.032) 0.040 (0.038)
Adj. R2 0.005 0.012 0.037 0.018 0.005 0.061 0.019 0.018 0.003 0.006 0.009
Log-likelihood 134.51 132.99 173.50 254.23 180.04 140.84 153.83 122.93 145.98 183.70 202.17
F-Statistic 0.75 0.40 3.06 2.55 2.57 4.31 0.06 3.47 0.83 0.74 0.55

Panel B: Multiple ex. VIX


Constant 0.017 (0.084) 0.038 (0.087) 0.080 (0.108) 0.336 (0.215) 0.045 (0.107) 0.099 (0.088) 0.010 (0.079) 0.176** (0.099) 0.067 (0.091) 0.295* (0.164) 0.152 (0.171)
ΔGSV(US) 0.015 (0.012) 0.005 (0.014) 0.035** (0.017) 0.042** (0.017) 0.028* (0.016) 0.047*** (0.014) 0.002 (0.018) 0.020** (0.010) 0.010 (0.014) 0.028* (0.015) 0.005 (0.021)
Cases 0.007 (0.019) 0.017 (0.018) 0.002 (0.024) 0.033 (0.052) 0.015 (0.024) 0.020 (0.017) 0.010 (0.018) 0.009 (0.014) 0.013 (0.019) 0.042 (0.033) 0.033 (0.030)
ΔEPU 0.006 (0.008) 0.018** (0.009) 0.017* (0.010) 0.022 (0.037) 0.005 (0.012) 0.011 (0.010) 0.004 (0.011) 0.002 (0.008) 0.017 (0.011) 0.012 (0.016) 0.000 (0.022)
ΔADS 0.184 (0.161) 0.109 (0.180) 0.737* (0.389) 1.582** (0.655) 0.556* (0.281) 0.433** (0.175) 0.387 (0.262) 0.015 (0.149) 0.119 (0.155) 0.544 (0.369) 0.217 (0.274)
ΔTERM 3.084* (1.638) 4.402*** (1.125) 2.466 (2.079) 3.749 (10.494) 0.816 (3.220) 0.463 (1.805) 0.352 (2.618) 2.049 (1.343) 1.190 (1.609) 1.867 (1.927) 2.664 (4.951)
ΔCSPRD 0.657 (1.244) 0.482 (1.305) 3.102* (1.807) 2.781 (3.253) 0.408 (1.899) 3.885*** (0.937) 0.516 (0.927) 2.548*** (0.853) 0.409 (1.588) 0.363 (1.441) 1.319 (2.770)
return(1) 0.064 (0.129) 0.069 (0.081) 0.125 (0.084) 0.034 (0.084) 0.040 (0.112) 0.041 (0.118) 0.064 (0.121) 0.196** (0.091) 0.066 (0.132) 0.097 (0.091) 0.126 (0.095)
Adj. R2 0.033 0.113 0.129 0.066 0.013 0.146 0.019 0.091 0.011 0.024 0.017
Log-likelihood 128.90 133.73 164.00 246.35 175.70 132.12 0.73 114.96 140.73 177.21 197.09
F-Statistic 1.51 2.84 3.14 2.03 1.89 3.47 0.20 2.45 0.84 1.35 0.76

Panel C: Multiple inc. VIX


Constant 0.041 (0.086) 0.066 (0.084) 0.148 (0.102) 0.243 (0.232) 0.011 (0.118) 0.148 (0.097) 0.021 (0.078) 0.223*** (0.082) 0.077 (0.091) 0.289 (0.167) 0.079 (0.167)
ΔGSV(US) 0.003 (0.016) 0.162 (0.011) 0.001 (0.039) 0.004 (0.045) 0.004 (0.020) 0.023* (0.013) 0.004 (0.021) 0.005 (0.012) 0.014 (0.014) 0.025 (0.020) 0.047 (0.034)
Cases 0.010 (0.019) 0.020 (0.017) 0.008 (0.022) 0.021 (0.050) 0.022 (0.025) 0.027 (0.017) 0.008 (0.019) 0.016 (0.014) 0.014 (0.019) 0.041 (0.036) 0.023 (0.029)
ΔVIX 0.015* (0.008) 0.018* (0.009) 0.047*** (0.016) 0.063** (0.030) 0.032* (0.017) 0.033*** (0.011) 0.007 (0.012) 0.036*** (0.006) 0.007 (0.009) 0.004 (0.018) 0.056** (0.023)
ΔEPU 0.007 (0.008) 0.018** (0.008) 0.013 (0.009) 0.016 (0.038) 0.003 (0.012) 0.014 (0.010) 0.004 (0.011) 0.001 (0.008) 0.016 (0.011) 0.011 (0.017) 0.005 (0.022)
ΔADS 0.133 (0.140) 0.051 (0.200) 0.564* (0.318) 1.354** (0.574) 0.446 (0.270) 0.306* (0.162) 0.415 (0.265) 0.112 (0.133) 0.142 (0.157) 0.560 (0.350) 0.018 (0.266)
ΔTERM 2.602 (1.714) 3.792*** (1.286) 0.516 (3.050) 5.623 (10.152) 0.026 (3.000) 1.347 (1.388) 0.135 (2.687) 1.011 (1.021) 1.384 (1.681) 2.003 (1.913) 4.609 (4.734)
ΔCSPRD 1.047 (1.182) 0.898 (1.161) 1.888 (1.539) 1.167 (3.054) 0.387 (1.625) 3.049** (1.003) 0.700 (1.055) 1.622* (0.827) 0.240 (1.614) 0.478 (1.448) 0.065 (2.565)
return(-1) 0.038 (0.124) 0.126 (0.096) 0.141* (0.077) 0.031 (0.086) 0.005 (0.116) 0.008 (0.100) 0.056 (0.118) 0.158** (0.076) 0.071 (0.131) 0.097 (0.092) 0.091 (0.101)
2
Adj. R 0.050 0.138 0.238 0.101 0.050 0.242 0.026 0.267 0.018 0.014 0.072
Log-likelihood 127.52 120.70 156.19 243.83 173.24 125.48 149.45 103.47 140.51 177.17 191.89
F-Statistic 1.67 3.03 4.94 2.43 1.66 5.04 0.68 5.60 0.08 1.19 1.97

Note(s): This table reports the estimated coefficients for the regression specified in Eqn (1). The dependent variable is the daily return on one of the 11 S&P sector indexes
(in excess of the risk-free 1M T-Bill rate and the S&P500 return).The key explanatory variables are the daily change (first difference) in the Google Search Volume (ΔGSV)
for “corono virus” in the US and the natural log of the daily number of new US cases of COVID-19. Control variables include the CBOE implied volatility index (VIX),
economic policy uncertainty (EPU), the Aruoba–Diebold–Scotti business conditions index (ADS), the 1–10 years term premium (TERM) and the spread between corporate
bonds yielded Baa and Aaa (CSPRD). Newey–West standard errors are shown in parentheses. ***, ** and * indicate statistical significance at the 1, 5 and 10% level
respectively. Sample period: 31 December 2019–31 May 2020
33
during the
attention

of investor attention on
COVID-19 crisis
Investor

sector)
equity returns (by
excess (Rs - (Rf þ Rm))
Table 5.
Regression: Influence
RBF Vozlyublennaia (2014) and Tantaopas et al. (2016) results for the Australian and New Zealand
13,1 markets (although contrary to their results for emerging markets).

3.5 Robustness test


The analysis reported so far uses the number of new COVID-19 cases as an explanatory
variable. Conceivably, investors may place more emphasis on the number of COVID-19–
34 related deaths rather than the number of cases. To test this, we repeat our earlier analysis with
the natural log of COVID-19 deaths in place of cases. Table 7 shows that the results are
qualitatively similar to those already reported. Panel A shows that the excess returns of all
sectors are negatively related to ΔGSV, and panel B shows that ΔGSV positively impacts
consumer staples and healthcare sectors relative to the overall market. For excess returns
(panel B), the effect of a 1 SD increase in GSV ranges from 1.08% (0.044 3 24.5) for energy
stocks to þ1.18% for the healthcare sector.
As was the cases with COVID-19 cases, deaths have a negative but statistically
insignificant impact on excess returns. Our results suggest that neither the current level of
cases or deaths are particularly important to investment decisions during the crisis period.
One explanation could be that information acquired from Google searches is more forward-
looking (e.g. possibility of vaccine or easing of lockdown measures) than the number of cases
or deaths and so more pertinent to the investment decision.

3.6 Global markets


The COVID-19 pandemic has global repercussions, and so we conclude our empirical analysis
by investigating the relationship between investor attention and stock sector returns at the
global level. Since our set of control variables is not available for global stimuli at a daily
frequency, our analysis is limited to the bivariate case. Table 8 reports the estimated
coefficients for returns adjusted for the 1M T-Bill rate (panel A) and returns adjusted for the
risk-free rate and overall market returns (Panel B). The results identified for the US are
reflected in the global results. Higher levels of retail investor attention are associated with
lower stock returns in all sectors, with the energy sector most influenced (panel A). – a 1 SD

F-statistic

ΔGSV(US) does not Granger cause rSP500 2.459**


rSP500 does not Granger cause ΔGSV(US) 1.761
ΔGSV(US) does not Granger cause ΔVIX 3.780***
ΔVIX does not Granger cause ΔGSV(US) 1.521
ΔVIX does not Granger cause rSP500 3.695***
rSP500 does not Granger cause ΔVIX 1.713
rSP500 does not Granger cause cases 1.440
Cases does not Grange cause rSP0 0.716
ΔVIX does not Granger cause cases 1.414
Cases does not Granger cause ΔVIX 1.072
ΔGSV(US) does not Granger cause Cases 1.281
Cases does not Granger cause ΔGSV(US) 1.213
Note(s): This table reports the F-statistics for tests of Granger causality between a selection of the key
variables in the study. This includes the daily return on the S&P500 Composite Index (in excess of the 1M T-Bill
rate), the daily change (first difference) in the Google Search Volume (ΔGSV) for “coronovirus” in the US, the
Table 6. natural log of the daily number of new US cases of COVID-19 and the CBOE implied volatility index (VIX).
Testing for Granger Estimates are provided for lag length of 5. ***, ** and * indicate statistical significance at the 1, 5 and 10% level,
causality respectively. Sample period: 31 December 2019–31 May 2020
S&P500 Comms Cons. Disc Cons. Staples Energy Financials Healthcare Industrials IT Materials Real estate Utilities

Panel A: Rs – Rf
Constant 0.229 (0.167) 0.168 (0.164) 0.246 (0.181) 0.054 (0.205) 0.158 (0.312) 0.287 (0.225) 0.095 (0.174) 0.289 (0.202) 0.429 (0.197) 0.204 (0.185) 0.602** (0.300) 0.306 (0.327)
ΔGSV(US) 0.117*** (0.038) 0.110 *** (0.033) 0.128*** (0.032) 0.113* (0.064) 0.104* (0.060) 0.129*** (0.044) 0.087** (0.037) 0.113** (0.048) 0.114*** (0.037) 0.130*** (0.039) 0.090** (0.042) 0.171*** (0.061)
Log_deaths(US) 0.025 (0.056) 0.001 (0.049) 0.009 (0.066) 0.012 (0.042) 0.040 (0.097) 0.080 (0.079) 0.013 (0.052) 0.059 (0.061) 0.041 (0.059) 0.017 (0.069) 0.111 (0.095) 0.062 (0.084)
Adj. R2 0.636 0.675 0.595 0.509 0.439 0.513 0.552 0.478 0.691 0.546 0.424 0.355
F-statistic 23.11 27.30 19.51 14.10 10.91 14.28 16.64 12.57 29.24 16.18 10.31 7.97

Comms Cons. Disc Cons. Staples Energy Financials Healthcare Industrials IT Materials Real estate Utilities

Panel B: Rs - (Rf þ Rm)


Constant 0.046 (0.080) 0.026 (0.092) 0.150 (0.152) 0.187 (0.269) 0.036 (0.138) 0.106 (0.097) 0.004 (0.100) 0.171* (0.101) 0.043 (0.102) 0.298* (0.161) 0.068 (0.192)
ΔGSV(US) 0.016 (0.012) 0.004 (0.014) 0.036** (0.016) 0.044** (0.017) 0.030* (0.016) 0.048*** (0.014) 0.001 (0.018) 0.020** (0.010) 0.010 (0.015) 0.026* (0.015) 0.004 (0.022)
Log_seaths(US) 0.018 (0.024) 0.022 (0.024) 0.015 (0.035) 0.013 (0.074) 0.044 (0.035) 0.032 (0.025) 0.016 (0.022) 0.013 (0.021) 0.014 (0.029) 0.064 (0.048) 0.027 (0.040)
Adjusted R2 0.037 0.112 0.130 0.064 0.020 0.148 0.019 0.091 0.013 0.027 0.022
F-statistic 1.55 2.82 3.16 1.99 1.30 3.51 0.74 2.45 0.81 1.40 0.69
Note(s): This table reports the estimated coefficients for the regression specified in Eqn (1). The dependent variable is the daily return on the S&P500 Composite index or
one of the 11 S&P sector indexes. In panel A, the return is computed as the excess of the risk-free 1M T-Bill rate only. In panel B, the return is computed as the excess of the
S&P500 return and the 1M T-Bill rate. The key explanatory variables are the daily change (first difference) in the Google Search Volume (ΔGSV) for “coronovirus” in the
US and the natural log of the daily number of US deaths from COVID-19. Control variables (coefficients not reported) include the economic policy uncertainty (EPU), the
Aruoba–Diebold–Scotti business conditions index (ADS), the 1–10 years term premium (TERM) and the spread between corporate bonds yielded Baa and Aaa (CSPRD).
Newey–West standard errors are shown in parentheses. ***, ** and * indicate statistical significance at the 1, 5 and 10% level, respectively. Sample period: 31 December
2019–31 May 2020
35
during the
attention

impact of COVID-19–
COVID-19 crisis
Investor

Robustness: Does the


Table 7.

related deaths differ


from cases?
36
13,1
RBF

Table 8.
Global stock returns
and investor attention
Panel A: Rs - Rf Panel B: Rs - (Rf þ Rm)
Constant ΔGSV(Global) Log_Cases Adj. R2 Constant ΔGSV(Global) Cases Adj. R2

MSCI World 0.119 (0.316) 0.160** (0.072) 0.016 (0.043) 0.126


Commodities 0.015 (0.263) 0.156** (0.066) 0.007 (0.034) 0.145 0.134 (0.098) 0.004 (0.013) 0.008 (0.013) 0.017
Consumer discretioanry 0.221 (0.350) 0.166** (0.068) 0.034 (0.049) 0.132 0.105 (0.081) 0.006 (0.021) 0.019 (0.012) 0.018
Consumer staples 0.063 (0.235) 0.101** (0.041) 0.006 (0.029) 0.054 0.056 (0.133) 0.069* (0.038) 0.009 (0.020) 0.119
Energy 0.854 (0.536) 0.250*** (0.080) 0.072 (0.068) 0.118 0.734 (0.295) 0.090** (0.042) 0.057 (0.038) 0.041
Financials 0.329 (0.381) 0.150* (0.089) 0.021 (0.056) 0.071 0.210 (0.112) 0.010 (0.020) 0.005 (0.020) 0.045
Healthcare 0.017 (0.277) 0.124** (0.054) 0.012 (0.034) 0.117 0.164 (0.081) 0.023* (0.012) 0.012 (0.014) 0.014
Industrials 0.283 (0.357) 0.137* (0.078) 0.028 (0.051) 0.100 0.102 (0.103) 0.016 (0.026) 0.004 (0.018) 0.001
IT 0.249 (0.342) 0.205** (0.083) 0.011 (0.044) 0.143 0.368*** (0.115) 0.045** (0.018) 0.027* (0.013) 0.042
Materials 0.438 (0.408) 0.128* (0.067) 0.053 (0.053) 0.094 0.319** (0.147) 0.033 (0.029) 0.037** (0.017) 0.019
Real estate 0.076 (0.377) 0.182* (0.105) 0.014 (0.059) 0.119 0.193 (0.197) 0.021 (0.042) 0.029 (0.028) 0.019
Utilities 0.213 (0.305) 0.133* (0.078) 0.023 (0.044) 0.069 0.332* (0.175) 0.027 (0.024) 0.039** (0.019) 0.015
Note(s): This table reports the estimated coefficients for the regression where the dependent variable is the daily return on the MSCI World index or one of the 11 MSCI
World sector indexes. The explanatory variables are the daily change (first difference) in the global Google Search Volume (ΔGSV) for “coronovirus” and the daily increase
in global COVID-19 cases reported to ECDC. Newey–West standard errors are shown in parentheses. ***, ** and * indicate statistical significance at the 1, 5 and 10% level,
respectively. Sample period: 31 December 2019 to31 May 2020
increase in GSV(global) is associated with a 6.75% (0.25 * 27.0) return. The results are Investor
consistent with those found in Da et al. (2015) and Chen (2017). attention
Once returns are adjusted for overall market performance (panel B); greater investor
attention leads to relative outperformance (underperformance) for consumer staples and
during the
healthcare (energy and IT). The effect of a 1 SD increase in GSV ranges from 2.43% for COVID-19 crisis
energy stocks to þ1.86% for the consumer staples sector. One difference is that, for a small
number of sectors (materials, utilities), the quantity of reported cases has a statistically
significant negative relationship with market-adjusted returns. Possible explanations include 37
supply chain disruptions, the highly globalized nature of the materials sector (which closely
relies on global commodity prices) and a fall in demand for the services provided by utilities as
the increase in cases forces workplaces to close [9].

4. Concluding Remarks
COVID-19 has had an immense impact on global stock markets, with no sector escaping its
effects. Investor attention towards COVID-19 surged as the virus spread and political
decisions imposed on everyday life. Our results show that the heightened attention of
investors, proxied by Google searches, negatively influenced US (and global) stock returns
and helps to explain the heterogeneity of returns across stock market sectors. Relatively
speaking, some sectors appear gaining from the increased attention, with outperformance
centred in the sectors most likely to benefit (or likely to lose least) from the crisis and
associated spending by households and government. This would suggest a rational action by
retail investors, consistent with the information discovery hypothesis, to the extent that they
are searching for information so as to comprehend the impact of COVID-19 on relative
financial performance of different stock sectors.

Notes
1. https://asic.gov.au/about-asic/news-centre/find-a-media-release/2020-releases/20-102mr-retail-
investors-at-risk-in-volatile-markets/
2. Source: https://data.worldbank.org/indicator/CM.MKT.LCAP.CD
3. The 11 GICS sectors are: Communications, consumer discretionary, consumer staples, energy,
financials, healthcare, industrials, I.T., materials, real estate and utilities.
4. https://trends.google.com/
5. Other studies (e.g. Da et al., 2015; Bijl et al., 2016) are required to normalize the GSV index over longer
sample periods since the frequency of GSV data decreases as the sample period increases.
6. We repeat our analysis using “COVID” and “COVID-19” search terms and achieve qualitatively
similar results.
7. https://www.ecdc.europa.eu/en/publications-data/download-todays-data-geographic-distribution-
covid-19-cases-worldwide
8. The data is used by respected resources such as the “Our World in Data” website supported by the
University of Oxford: https://ourworldindata.org/coronavirus-source-data
9. See: https://www.ibisworld.com/industry-insider/coronavirus-insights/coronavirus-update-
industry-fast-facts

References
Al-Awadhi, A.M., Alsaifi, K., Al-Awadhi, A. and Alhammadi, S. (2020), “Death and contagious
infectious diseases: impact of the COVID-19 virus on stock market returns”, Journal of
Behavioral and Experimental Finance, Vol. 27, p. 100326.
RBF Ali, M., Alam, N. and Rizvi, S.A.R. (2020), “Coronavirus (COVID-19) – an epidemic or pandemic for
financial markets”, Journal of Behavioral and Experimental Finance, Vol. 27, p. 100341.
13,1
Andrei, D. and Hasler, M. (2014), “Investor attention and stock market volatility”, Review of Financial
Studies, Vol. 28, pp. 33-72.
Aouadi, A., Arouri, M. and Teulon, F. (2013), “Investor attention and stock market activity: evidence
from France”, Economic Modelling, Vol. 35, pp. 674-681.
38 Baig, A.S., Butt, H.A., Haroon, O. and Rizvi, S.A.R. (2020), “Deaths, panic, lockdowns and US equity
markets: the case of COVID-19 pandemic”, Working Paper, SSRN: 3584947.
Baker, S.R., Bloom, N., Davis, S.J., Kost, K., Sammon, M. and Viratyosin, T. (2020), “The unprecedented
stock market reaction to COVID-19”, The Review of Asset Pricing Studies, Vol. 10, pp. 742-758.
Bank, M., Larch, M. and Peter, G. (2011), “Google search volume and its influence on liquidity and
returns of German stocks”, Financial Markets and Portfolio Management, Vol. 25, pp. 239-264.
Barber, B.M. and Odean, T. (2008), “All the glitters: the effect of attention and news on the buying
behaviour of individual and institutional investors”, Review of Financial Studies, Vol. 21,
pp. 785-818.
Barber, B.M. and Odean, T. (2013), “The behaviour of individual investors”, in Constantinides, G.M.,
Harris, M. and Stulz, R.M. (Eds), Handbook of the Economics of Finance, Elsevier, Oxford, Vol. 2,
pp. 1533-1570.
Bijl, L., Kringhaug, G., Molnar, P. and Sandvik, E. (2016), “Google searches and stock returns”,
International Review of Financial Analysis, Vol. 45, pp. 150-156.
Chen, T. (2017), “Investor attention and global stock returns”, Journal of Behavioral Finance, Vol. 18,
pp. 358-372.
Cheng, C.M., Huang, A.Y. and Hu, M.C. (2019), “Investor attention and stock price movement”, Journal
of Behavioural Finance, Vol. 20, pp. 294-303.
Da, Z., Engelberg, J. and Gao, P. (2011), “In search of attention”, Journal of Finance, Vol. 66,
pp. 1461-1499.
Da, Z., Engelberg, J. and Gao, P. (2015), “The sum of all FEARS investor sentiment and asset prices”,
Review of Financial Studies, Vol. 28, pp. 1-32.
Dimpfl, T. and Jank, S. (2015), “Can internet search queries help to predict stock market volatility?”,
European Financial Management, Vol. 22, pp. 171-192.
Ding, R. and Hou, W. (2015), “Retail investor attention and stock liquidity”, Journal of International
Financial Markets, Institutions and Money, Vol. 37, pp. 12-26.
Garcia, D. (2013), “Sentiment during recessions”, Journal of Finance, Vol. 68, pp. 1267-1300.
Goddard, J., Kita, A. and Wang, Q. (2015), “Investor attention and FX Market volatility”, Journal of
International Financial Markets, Institutions and Money, Vol. 38, pp. 79-96.
Granger, C.W.J. (1969), “Investigating causal relations by econometric models and cross-spectral
methods”, Econometrica, Vol. 37, pp. 424-438.
Han, L., Wu, Y. and Yin, L. (2018), “Investor attention and currency performance: international
evidence”, Applied Economics, Vol. 50, pp. 2525-2551.
Haroon, O. and Rizvi, S.A.R. (2020), “COVID-19: media coverage and financial markets behaviour – a
sectoral inquiry”, Journal of Behavioral and Experimental Finance, Vol. 27, p. 100343.
Heyman, D., Lescrauwaet, M. and Stieperaere, H. (2019), “Investor attention and short-term return
reversals”, Finance Research Letters, Vol. 29, pp. 1-6.
Hirshleifer, D. (2015), “Behavioral finance”, Annual Review of Financial Economics, Vol. 7, pp. 133-159.
Huberman, G. and Regev, T. (2001), “Contagious speculation and a cure for cancer: a nonevent that
made stock prices soar”, Journal of Finance, Vol. 56, pp. 387-396.
Kahneman, D. (1973), Attention and Effort, Prentice-Hall, Englewood Cliffs, NJ.
Kim, N., Lucivjanska, K., Molnar, P. and Villa, R. (2019), “Google searches and stock market activity: Investor
evidence from Norway”, Finance Research Letters, Vol. 28, pp. 208-220.
attention
Lim, S.S. and Teoh, S.H. (2010), “Limited attention”, in Baker, H.K. and Nofsinger, J.R. (Eds), Behavioral
Finance, Wiley & Sons, New Jersey, pp. 295-312.
during the
Odean, T. (1999), “Do investors trade too much?”, American Economic Review, Vol. 89, pp. 1279-1298.
COVID-19 crisis
Peng, L. and Xiong, W. (2006), “Investor attention, overconfidence and category learning”, Journal of
Financial Economics, Vol. 80, pp. 563-602. 39
Preis, T., Reith, D. and Stanley, H.E. (2010), “Complex dynamics of our economic life on different
scales: insights from search engine query data”, Philosophical transactions of The Royal Society
A, Vol. 368, pp. 5707-5719.
Ramelli, S. and Wagner, A.F. (2020), “Feverish stock price reactions to COVID-19”, Review of
Corporate Finance Studies, Vol. 9, pp. 622-655.
Ru, H., Yang, E. and Zou, K. (2020), “What do we learn from SARS-CoV-1 to SARS-CoV-2: evidence
from global stock markets”, Working Paper.
Schell, D., Wang, M. and Huynh, T.L.D. (2020), “This time is indeed different: a study on global market
reactions to public health crisis”, Journal of Behavioral and Experimental Finance, Vol. 27,
100349.
Smales, L.A. (2020), “Investor attention and global market returns during the COVID-19 crisis”,
International Review of Financial Analysis, Vol. 73, 101616.
Smith, G.P. (2012), “Google internet search activity and volatility prediction in the market for foreign
currency”, Finance Research Letters, Vol. 9, pp. 103-110.
Smith, C.A. and Ellsworth, P.C. (1985), “Patterns of cognitive appraisal in emotion”, Journal of
Personality and Social Psychology, Vol. 48, pp. 813-838.
Takeda, F. and Wakao, T. (2014), “Google search intensity and its relationship with returns and
trading volume of Japanese stocks”, Pacific-Basin Finance Journal, Vol. 27, pp. 1-18.
Tang, W. and Zhu, L. (2017), “How security prices respond to a surge in investor attention: evidence
from Google search of ADRs”, Global Finance Journal, Vol. 33, pp. 38-50.
Tantaopas, P., Padungsaksawasdi, C. and Treepongkaruna, S. (2016), “Attention effect via internet
search intensity in Asia-Pacific stock markets”, Pacific-Basin Finance Journal, Vol. 38,
pp. 107-124.
Vlastakis, N. and Markellos, R.N. (2012), “Information demand and stock market volatility”, Journal of
Banking and Finance, Vol. 36, pp. 1808-1821.
Vozlyublennaia, N. (2014), “Investor attention, index performance, and return predictability”, Journal
of Banking and Finance, Vol. 41, pp. 17-35.
Whaley, R.E. (2009), “Understanding the VIX”, Journal of Portfolio Management, Vol. 35, pp. 98-105.
Zhang, D., Hu, M. and Ji, Q. (2020), “Financial markets under the global pandemic of COVID-19”,
Finance Research Letters, Vol. 36, 101528.

Corresponding author
Lee A. Smales can be contacted at: lee.smales@uwa.edu.au

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

You might also like